The invention relates to the field of buried channel strained-Si FETs, and in particular to these FETs using a supply layer created through ion implantation.
The advent of relaxed SiGe alloys on Si substrates introduces a platform for the construction of new Si-based devices. These devices have the potential for wide application due to the low cost of using a Si-based technology, as well as the increased carrier mobility in strained layers deposited on the relaxed SiGe.
As with most new technologies, implementing these advances in a Si CMOS fabrication facility requires additional innovation. For example, some of the potential new devices are more easily integrated into current Si processes than other devices. Since process technology is directly relevant to architecture, particular innovations in process technology can allow the economic fabrication of new applications/architectures.
FIGS. 1A and 1B are schematic block diagrams showing the variety of strained Si devices that are possible to fabricate given the advent of relaxed SiGe buffer layers. FIG. 1A shows a surface channel strained Si MOSFET 100. In this configuration, a tensile, strained Si channel 102 is deposited on relaxed SiGe layer 104 with a Ge concentration in the range of 10–50%. This relaxed SiGe layer is formed on a Si substrate 108 through the use of a compositionally graded SiGe buffer layer 106. A conventional MOS gate stack 110 is on the strained silicon channel and consists of an oxide layer 112, a poly-Si electrode 114, and a metal contact layer 116. Doped source 118 and drain 120 regions are also formed on either side of the gate stack to produce the MOSFET device structure.
A buried channel strained Si high electron mobility transistor (HEMT) 130 is shown in FIG. 1B. In this configuration, the strained Si 102 atop the relaxed SiGe 104 has been capped with a thin SiGe cap layer 132. The strained Si layer generally has a thickness between 2–30 nm, while the SiGe cap layer has a thickness between 2–20 nm. A metal Schottky gate 134 on the SiGe cap layer is commonly used on the HEMT, and, as in the MOSFET structure, doped source 118 and drain 120 regions are formed on each side of this gate.
FIG. 1C shows a buried channel strained Si MOSFET 140. This device has the same Si/SiGe layer structure as the HEMT configuration, but with a full MOS gate stack 142, consisting of oxide 144, poly-Si 146, and metal 148 layers, rather than the metal Schottky gate.
It is important to separate these devices into two categories, surface channel devices, of which an embodiment is shown in FIG. 1A, and buried channel devices, of which embodiments are shown in FIGS. 1B and 1C. In the case of the surface channel device, a light background doping in the SiGe during epitaxial growth or by implantation is sufficient to position the Fermi level such that a MOSFET constructed from the strained surface channel has reasonably large threshold values. Thus, the surface can be inverted for either p or n channel operation.
FIGS. 2A and 2B are the energy band diagram for the case of the surface channel FET for an NMOS device, (A) at zero bias, and (B) at a bias to turn on the transistor, respectively. When the transistor is turned on, a relatively large electric field exists in the normal direction to the surface plane, and the electrons are attracted to the surface and operate in the strained Si surface channel. The speed of the transistor is increased due to the fact that the electrons reside in the high mobility, strained Si surface channel. However, the device has noise performance similar to a conventional Si MOSFET since the carriers scatter off the SiO2/Si interface, and the device, although it possesses a mobility larger than that of a conventional Si device, still has a mobility that is limited by the SiO2/Si interface.
However, it is known from III–V materials that a buried channel device should possess a much higher electron mobility and lower noise performance. For example, the structures shown in FIG. 1B and C should have higher channel mobility and lower noise performance than the device in FIG. 1A since the electron scatters off a semiconductor interface instead of an oxide interface.
A crucial flaw in the device shown in FIG. 1C that leads to processing difficulties and limitations in circuit layout and architectures is that when the device is biased to invert the channel and turn the device on, the band structure is such that many of the carriers leave the buried channel. FIG. 3 is an energy band diagram showing schematically the problem with a buried channel device in which there is no dopant supply layer. The field required to turn on the device empties the buried channel. This effectively creates a surface channel device even though the buried channel layer is present in the heterostructure.
The applied gate bias of FIG. 3 has bent the bands such that many of the electrons from the well escape confinement and create an inversion layer at the oxide/semiconductor interface. Since transconductance of a field effect device is high if the mobility and the number of carriers is high, a high performance FET, i.e., even higher performance than the surface channel device, is difficult to achieve. At low vertical fields, the electrons are in the high mobility buried channel, but there are few in number. If the device is turned on and inverted as shown in FIG. 3, the carrier density in the surface channel is high but the mobility is reduced since the carriers are now at the rough oxide interface.
One way to solve this problem is to insert a dopant supply layer into the structure, as shown in FIG. 4A. FIG. 4A is a schematic block diagram of a structure 400 in which the buried channel can be occupied with a high density of electrons via the insertion of a layer of donor atoms. It will be appreciated that an equivalent schematic can be constructed for a buried hole channel with a layer of acceptor atoms.
The structure 400 includes a strained Si channel 402 positioned between two SiGe layers, a relaxed SiGe layer 404 and a thin SiGe cap layer 406. Although FIG. 4A shows a dopant supply layer 408 in the SiGe cap, the dopants can be introduced into either SiGe layer. As has been shown in the III–V buried channel devices, this layer configuration creates a band structure where now the buried channel is occupied, as shown in FIG. 4B. In this figure, the supply layer leads to localized band bending and carrier population of the buried strained Si. In the strained Si, the conduction band has been lowered beneath the Fermi level, resulting in a high carrier density in the high mobility channel. One disadvantage of this structure is that now the transistor is on without any applied voltage, and a voltage is supplied to the gate to turn off the transistor. Thus, this transistor is normally on or depletion-mode. As a result, the device is useful in analog and logic applications, but is not easily implemented in a conventional CMOS architecture.
Common accepted practice in the buried channel heterostructure FETs is to use a dopant supply layer that is introduced in an epitaxial step, i.e., deposited during the epitaxial process that creates the Si/SiGe device structure. This dominant process originates from the III–V research device community (AlGaAs/GaAs materials system). However, this epitaxial dopant supply layer is undesirable since it reduces thermal budget and limits the variety of devices available in the circuit. For example, if the dopant supply layer is introduced in the epitaxial step, when processing begins, the thermal budget is already constrained due to diffusion of the supply layer dopants. All devices in the circuit must also now be buried channel devices with similar thresholds, since any removal of the dopant layer in a particular region would require complete etching of the local area and removal of critical device regions.